摘要
通过分析小波变换的多相表示和GPU通用计算模型,结合现代GPU的多纹理和多渲染目标特性,提出了一种基于GPU与多相表示的二维离散小波变换计算方法,该方法使小波变换的计算形式完全适合GPU的SIMD结构,同时大幅减少了纹理访问次数,充分利用了GPU的矢量运算和二维寻址能力,实验表明该方法在处理速度上有大幅的提高。
By analyzing the DWT's polyphase representation and GPU's general-purpose computation model, this study presents a 2D DWT algorithm based on GPU and polyphase, which combines with modem CPU's multi-texture property and multiple render targets property. The computing form is well suited to GPU's SIMD architecture in the method presented. The number of texture visit is significantly reduced. The GPU' s vector processing ability and 2D address accessing ability is well utilized. The experimental result shows that the processing speed is greatly improved.
出处
《国防科技大学学报》
EI
CAS
CSCD
北大核心
2011年第3期111-114,共4页
Journal of National University of Defense Technology
基金
国家部级科研项目